CN103679631A - Method for amplifying images - Google Patents

Method for amplifying images Download PDF

Info

Publication number
CN103679631A
CN103679631A CN201210346575.XA CN201210346575A CN103679631A CN 103679631 A CN103679631 A CN 103679631A CN 201210346575 A CN201210346575 A CN 201210346575A CN 103679631 A CN103679631 A CN 103679631A
Authority
CN
China
Prior art keywords
pixel
image
class
pixels
block
Prior art date
Application number
CN201210346575.XA
Other languages
Chinese (zh)
Other versions
CN103679631B (en
Inventor
干宗良
朱秀昌
夏青
张园园
石腾
Original Assignee
华为技术有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 华为技术有限公司 filed Critical 华为技术有限公司
Priority to CN201210346575.XA priority Critical patent/CN103679631B/en
Publication of CN103679631A publication Critical patent/CN103679631A/en
Application granted granted Critical
Publication of CN103679631B publication Critical patent/CN103679631B/en

Links

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING; COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T3/00Geometric image transformation in the plane of the image
    • G06T3/40Scaling the whole image or part thereof
    • G06T3/4007Interpolation-based scaling, e.g. bilinear interpolation

Abstract

The invention provides a method for amplifying images so as to solve the problem that the amplifying speed of the images is affected by high calculation complexity in the parameter construction process of a non-local filter. The method comprises the following steps: dividing the pixel points of low-resolution and hr (high-resolution) images into a low-resolution (lr) pixel class and an hr pixel class according to pixel characteristics; determining the pixel points which are edge pixel points and belong to the hr pixel class according to the edge pixel points, respectively calculating similar distances between all the determined hr pixel class of pixel points and similar distances between the lr pixel class of pixel points, and obtaining the pixel points of which the similar distances meet similarity in the lr pixel class; constructing the parameters of the non-local (nl) filter. The parameters of the nl filter are generated only through the calculation of the similar distances between the hr pixel class of pixel points and the similar distances between the lr pixel class of pixel points, and the calculation complexity of the parameters of the nl filter is simplified.

Description

A kind of method of enlarged image

Technical field

The invention belongs to computer realm, relate to a kind of enlarged image method and device thereof.

Background technology

Along with the development of multimedia technology and computer networking technology, image information plays an increasingly important role in people's work, studying and living.Image resolution ratio is a kind of tolerance of imaging system to output image details resolution characteristic, and resolution is higher means that picture element density is higher, and details is meticulousr, and the information providing is also abundanter.Yet due to the restriction of the physical attribute of Image-forming instrument own and the interference of external environmental condition, image resolution ratio often can not meet the requirement of some application scenarios, in 3D Video coding, be subject to limit on transmission bandwidth, conventionally depth map is dwindled in service end, in terminal, amplify, this has wherein used the resolution that image amplifying technique improves image.Other use scenes is also very high to real-time requirement, and the image of for example video conference, video calling amplifies, if can not meet real-time requirement, can affect user's experience, therefore needs image amplifying technique fast in order to be effective, to improve speed.

Early stage image magnification method mostly can not be gone back comparatively responsive edge and the texture of human eye in original image preferably.In above-mentioned 3D Video coding, image can be regarded as to a kind of signal, and the error the image after amplifying and original depth map between the source figure before service end is dwindled is considered as a kind of noise, therefore can in image amplification process, add filtering to remove noise.

Non local (non-local, nl) wave filter is the current generally acknowledged best wave filter of denoising effect.Had at present in some images amplifications and carried out denoising with non local wave filter, wherein with the non local wave filter based on interpolation algorithm, be most widely used again, but but have the problem that calculated amount is high in non local filter parameter construction process.

Summary of the invention

The present invention amplifies speed issue for solving because non local filter parameter construction process computation complexity height has affected image:

First aspect, the embodiment of the present invention provides a kind of image magnification method, the method comprises: low resolution (lr) image is generated to high resolving power (hr) image by image interpolation algorithm, described hr image is carried out to the pixel that rim detection is obtained edge in image; By the pixel of described hr image according to pixels characteristic be divided into lr pixel class and hr pixel class, described lr pixel class comprises the pixel coming from lr image, described hr pixel class comprises the pixel calculate generating through interpolation algorithm; According to the pixel at described edge, determine to be the pixel at edge and the pixel that belongs to hr pixel class, calculate respectively the pixel of described each definite hr pixel class and the similarity distance of the pixel in described lr pixel class, and obtain the pixel that similarity distance in lr pixel class meets similarity; According to the pixel in the pixel in described each definite hr pixel class and the corresponding lr pixel class that meets similarity, construct the parameter of non local nl wave filter; Thereby utilize described nl wave filter to filter through the initial enlarged image of interpolator arithmetic and obtain final enlarged image.

In the possible implementation of the first of first aspect, when the enlargement factor of described interpolation algorithm is N, wherein, N is greater than 1 natural number, described by described hr image according to pixels characteristic be divided into lr pixel class and hr pixel class, specifically comprise: by calculate the pixel coming from lr image in the hr image generating through interpolation algorithm, be classified as first group of lr pixel class, and generate first group of lr pixel class image; Each pixel in first group of lr pixel class of take is reference point, by predetermined division rule, hr image is divided into the block of pixels of a plurality of N * N, and the pixel that the block of pixels of each N * N is had to identical position relationship with respect to described reference point is assigned in same group of hr pixel class, total N2-l group hr pixel class in described hr image, generates N2-l group hr pixel class image; Each pixel in described first group of lr pixel class image and N2-l group hr pixel class image has respectively coordinate and the coordinate in hr image in the pixel class image under separately.

In conjunction with the possible implementation of the first of first aspect, in the possible implementation of the second, it is the pixel at edge and the pixel that belongs to hr pixel class that the pixel at the described edge of described foundation is determined, wherein, when specifically determining coordinate, be (X, the pixel at edge Y) meets while belonging to hr pixel class, calculate the pixel (X of described definite hr pixel class, the process of the similarity distance of the pixel Y) and in described lr pixel class, specifically comprise: according to the pixel coordinate (X at described edge, Y), at judgement (X, Y) belong to after the pixel in hr pixel class, determine that described pixel specifically belongs to the hr pixel class image of k group, and the coordinate in the hr pixel class image of described k group is (xk, yk), wherein k ∈ [2, N2], calculate respectively the similarity distance of the pixel that in the hr pixel class image of each pixel and described k group in first group of lr pixel class image, coordinate is (xk, yk).

In conjunction with the possible implementation of the second of first aspect, in the third possible implementation, in specifically calculating first group of lr pixel class image, coordinate is (x1, y1) in pixel and k group hr pixel class image, coordinate is (xk, during the similarity distance of pixel yk), be specially: obtain in hr image with (X1, Y1) the first block of pixels centered by and (Xk, Yk) the second block of pixels centered by, (X1 wherein, Y1) be that in first group of lr pixel class image, coordinate is the pixel correspondence of (x1, the y1) coordinate in hr image; Wherein (Xk, Yk) is pixel correspondence that in k group hr pixel class image, coordinate is (xk, the yk) coordinate in hr image; All pixels that belong to first group of lr pixel class obtain described the first block of pixels in first group of lr pixel class image in, and be classified as the 3rd block of pixels; All pixels that belong to k group hr pixel class obtain described the second block of pixels in k group hr pixel class image in, and be classified as the 4th block of pixels; All pixels that belong to first group of lr pixel class obtain described the second block of pixels in first group of lr pixel class image in, and be classified as the 5th block of pixels; All pixels that belong to k group hr pixel class obtain described the first block of pixels in k group hr pixel class image in, and be classified as the 6th block of pixels; Calculate the 3rd block of pixels and the 4th block of pixels absolute error and obtain the first absolute error and, and calculate the 5th block of pixels and the 6th block of pixels absolute error and obtain the second absolute error and, by the first absolute error and with the second absolute error and addition, obtaining coordinate in first group of lr pixel class image is (x1, y1) in pixel and k group hr pixel class image, coordinate is the similarity distance between (xk, yk) pixel.

The first implementation in conjunction with first aspect, the second implementation or the third implementation, in the 4th kind of possible implementation, at described enlargement factor N, it is 2 o'clock, it is described that to take each pixel in first group of lr pixel class be reference point, by predetermined division rule, hr image is divided into the block of pixels of a plurality of N * N, and the pixel that the block of pixels of each N * N is had to identical position relationship with respect to described reference point is assigned in same group of hr pixel class, total N2-l group hr pixel class in described hr image, and generate N2-l group hr pixel class image, specifically comprise: each pixel in first group of lr pixel class of take is reference point, in the direction of the clock described hr image is divided into each block of pixels of 2 * 2, in each 2 * 2 block of pixels of dividing out based on described reference point, so that described reference point is for referencial use, successively the pixel of its excess-three hr pixel class except described reference point in described each 2 * 2 block of pixels is belonged to respectively to second group of hr pixel class, the 3rd group of hr pixel class and the 4th group of hr pixel class in the direction of the clock, respectively according to described first group of lr pixel class, second group of hr pixel class, the 3rd group of hr pixel class and the 4th group of corresponding first group of lr pixel class image, second group of hr pixel class image, the 3rd group of hr pixel class image and the 4th group of hr pixel class image of pixel generation that hr pixel class comprises.

In conjunction with the third implementation or the 4th kind of implementation of first aspect, in the 5th kind of implementation, described the 3rd block of pixels and the 4th block of pixels size are all M * M, the absolute error of described calculating the 3rd block of pixels and the 4th block of pixels and, specifically comprise: the pixel value that obtains each pixel in the 3rd block of pixels is v1, v2, v3, ..., in vn (M * M=n) and the 4th block of pixels, the pixel value of corresponding pixel points is u1, u2, u3, ..., un; And by formula S AD=|u1-v1|+|u2-v2|+|u3-v3|+......+|un-vn| calculate the 3rd block of pixels and the 4th block of pixels absolute error and.

In conjunction with first aspect or in conjunction with above-mentioned any one possible implementation, in the 6th kind of possible implementation, the pixel of each hr pixel class of the described calculative determination of described correspondence, obtain the pixel that similarity distance in lr pixel class meets similarity, specifically comprise: the pixel of each hr pixel class of corresponding described calculative determination, obtain P pixel of similarity distance minimum in lr pixel class as the pixel that meets similarity, wherein P is greater than 1 natural number.

In conjunction with above-mentioned any one possible implementation, in the 7th kind of possible implementation, pixel in pixel in described each the definite hr pixel class of described foundation and the corresponding lr pixel class that meets similarity is constructed the parameter of non local nl wave filter, specifically comprise: according to first pixel in described definite hr pixel class and corresponding described first pixel, meet pixel in P lr pixel class of similarity as parameter, calculate the weights of the pixel of described P lr pixel class; The weights of the pixel of the coordinate of the pixel correspondence of preserving the coordinate of described first pixel correspondence in hr image, a described P lr pixel class in hr image and described P lr pixel class; According to the method for first pixel in described hr pixel class, handle remaining pixel in described each definite hr pixel class.

The 7th kind of possible implementation in conjunction with first aspect, in the 8th kind of possible implementation, thereby the described nl of utilization wave filter filters through the initial enlarged image of interpolator arithmetic and obtains final enlarged image, specifically comprises: the weights of the pixel of the coordinate of the pixel correspondence of obtaining the coordinate of described first pixel of preserving in the parameter of nl wave filter, a described P lr pixel class in hr image and described P lr pixel class; The coordinate of pixel correspondence based on described P lr pixel class in hr image obtains P pixel value corresponding in initial enlarged image, and the weights of the pixel of described P lr pixel class of foundation obtain weighing computation results; Described weighing computation results is replaced to the pixel value of corresponding described first pixel in initial enlarged image; Pixel according to other hr pixel classes of preserving in described first pixel method successively parameter based on described nl wave filter is processed initial enlarged image.

The 8th kind of possible implementation in conjunction with first aspect, in the 9th kind of possible implementation, the coordinate of the described pixel correspondence based on described P lr pixel class in hr image obtains P pixel value corresponding in initial enlarged image, and the weights of the pixel of described P lr pixel class of foundation obtain weighing computation results, specifically comprise: the coordinate that meets the pixel of similarity according to the P preserving in the parameter of nl wave filter and described first pixel, obtain and amplify the pixel value H (i) on a described P coordinate in rear image, i=1, 2, ..., P, and utilize formula calculate weighing computation results, wherein represent that current high-resolution position pixel participates in the weights that weighted sum is used, it is the maximal value of getting in P weights above, wherein ω (i) is the weights of the pixel of P lr pixel class, and the big or small n of block of pixels in the parametric t by analogue noise attenuation degree and hr image, according to formula i=1 wherein, 2 ..., P calculates acquisition.

A second aspect of the present invention, a kind of enlarged image device is provided, comprise: pixel sort module, for by the pixel of described hr image according to pixels characteristic be divided into lr pixel class and hr pixel class, described lr pixel class comprises the pixel coming from lr image, and described hr pixel class comprises through interpolation algorithm calculates the pixel generating; Similarity distance computing module, determines to be the pixel at edge and the pixel that belongs to hr pixel class for the pixel according to described edge, respectively the pixel of each hr pixel class and the similarity distance of the pixel in described lr pixel class of calculative determination; Similitude acquisition module, meets the pixel of similarity for obtaining lr pixel class similarity distance; The parameter generation module of non local nl wave filter, for constructing the parameter of non local nl wave filter according to the pixel in the pixel of described each definite hr pixel class and the corresponding lr pixel class that meets similarity; Filtering module, thus for utilizing nl wave filter to filter through the initial enlarged image of interpolator arithmetic, obtain final enlarged image.

In the possible implementation of the first of second aspect, described pixel sort module, also comprise: pixel class image generates submodule, for the pixel that the hr image that calculates generation through interpolation algorithm is come from lr image, be classified as first group of lr pixel class, and generate first group of lr pixel class image; Each pixel in first group of lr pixel class of take is reference point, by predetermined division rule, hr image is divided into the block of pixels of a plurality of N * N, and the pixel that the block of pixels of each N * N is had to identical position relationship with respect to described reference point is assigned in same group of hr pixel class, total N2-1 group hr pixel class in described hr image, and generate N2-1 group hr pixel class image; Wherein, the enlargement factor that N is described interpolation algorithm, N is greater than 1 natural number; Each pixel in described first group of lr pixel class image and N2-1 group hr pixel class image all has respectively coordinate and the coordinate in hr image in pixel class image separately.

In the possible implementation of the second of second aspect, described similarity distance computing module also comprises mapping generation module, for ought specifically calculating first group of lr pixel class image coordinate, be (x1, y1) in pixel and k group hr pixel class image, coordinate is (xk, during the similarity distance of pixel yk), be specially: obtain in hr image with (X1, Y1) the first block of pixels centered by and (Xk, Yk) the second block of pixels centered by, (X1 wherein, Y1) be that in first group of lr pixel class image, coordinate is the pixel correspondence of (x1, the y1) coordinate in hr image; Wherein (Xk, Yk) is pixel correspondence that in k group hr pixel class image, coordinate is (xk, the yk) coordinate in hr image; All pixels that belong to first group of lr pixel class obtain described the first block of pixels in first group of lr pixel class image in, and be classified as the 3rd block of pixels; All pixels that belong to k group hr pixel class obtain described the second block of pixels in k group hr pixel class image in, and be classified as the 4th block of pixels; All pixels that belong to first group of lr pixel class obtain described the second block of pixels in first group of lr pixel class image in, and be classified as the 5th block of pixels; All pixels that belong to k group hr pixel class obtain described the first block of pixels in k group hr pixel class image in, and be classified as the 6th block of pixels; Described similarity distance computing module, also for calculate the 3rd block of pixels and the 4th block of pixels absolute error and obtain the first absolute error and, and calculate the 5th block of pixels and the 6th block of pixels absolute error and obtain the second absolute error and, by the first absolute error and with the second absolute error and addition, obtaining coordinate in first group of lr pixel class image is (x1, y1) in pixel and k group hr pixel class image, coordinate is the similarity distance between (xk, yk) pixel.

The present invention is by the method and apparatus of above-mentioned a kind of enlarged image, pixel characteristic after amplifying according to interpolation algorithm in image is divided into high-resolution pixel class image and low-resolution pixel class image, only by calculating the similarity distance of the pixel of high-resolution pixel class and high-resolution pixel class, generate the parameter of nl wave filter, simplified the computation complexity of nl filter parameter.

Accompanying drawing explanation

In order to be illustrated more clearly in the technical scheme of the embodiment of the present invention, to the accompanying drawing of required use in the embodiment of the present invention be briefly described below, apparently, below described accompanying drawing be only some embodiments of the present invention, for those of ordinary skills, do not paying under the prerequisite of creative work, can also obtain according to these accompanying drawings other accompanying drawing.

Fig. 1 is the schematic flow sheet of method of a kind of enlarged image of the embodiment of the present invention;

Fig. 2 is the lr pixel class of the embodiment of the present invention and the schematic flow sheet that hr pixel class is divided;

Fig. 3 be the embodiment of the present invention in enlargement factor, be the structure schematic flow sheet of the nl filter parameter of 2 o'clock;

Fig. 4 is the schematic flow sheet of method of a kind of enlarged image of the embodiment of the present invention;

Fig. 5 is the structural representation of a kind of enlarged image device of the embodiment of the present invention;

Fig. 6 is the structural representation of a kind of enlarged image device of the embodiment of the present invention;

Fig. 7 is the schematic diagram that in the direction of the clock hr image is divided into a plurality of 2 * 2 block of pixels in the present invention;

Fig. 8 calculates pixel similarity schematic diagram in hr image in the present invention;

Fig. 9 calculates pixel similarity correspondence to show the schematic diagram that is related in each pixel class image in hr image in the present invention;

Figure 10 is that the pixel correspondence of calculating similarity in the present invention in hr image is related to schematic diagram in pixel class image;

Figure 11 is the relation structure diagram of a kind of nl filter parameter of the embodiment of the present invention.

Embodiment

Below in conjunction with the accompanying drawing in the embodiment of the present invention, the technical scheme in the embodiment of the present invention is clearly and completely described, obviously, described embodiment is a part of embodiment of the present invention, rather than whole embodiment.Embodiment based in the present invention, the every other embodiment that those of ordinary skills obtain under the prerequisite of not making creative work, should belong to the scope of protection of the invention.

In the present invention, low-resolution image is exactly original image, high-definition picture is to compare the image of original image after interpolation algorithm amplifies, wherein low resolution and high resolving power, as just the restriction of both differences, are not the specific (special) requirements for the resolution sizes of image.

A kind of enlarged image embodiment of the method 100 provided by the invention, the present embodiment method is based on by low resolution (low resolution, lr) image generates high resolving power (high resolution by image interpolation algorithm, hr) image, described hr image is carried out to rim detection to be obtained on the basis of the pixel at edge in image and completes, as shown in Figure 1, process is as follows:

S101, by the pixel of described hr image according to pixels characteristic be divided into lr pixel class and hr pixel class, described lr pixel class comprises the pixel coming from lr image, described hr pixel class comprises the pixel calculate generating through interpolation algorithm.

Described pixel characteristic is specially: the pixel comprising in the described hr image generating through interpolation algorithm is the pixel coming from lr image, or calculates by interpolation algorithm the interpolating pixel point generating.Therefore described lr pixel class can be described as pixel in lr image and forms.

When the enlargement factor of described interpolation algorithm is N, wherein, N is greater than 1 natural number; Preferably, described by described hr image according to pixels characteristic be divided into lr pixel class and hr pixel class, specifically can be implemented as:

By calculate the pixel coming from lr image in the hr image generating through interpolation algorithm, be classified as first group of lr pixel class, and generate first group of lr pixel class image; Each pixel in first group of lr pixel class of take is reference point, by predetermined division rule, hr image is divided into the block of pixels of a plurality of N * N, and the pixel that the block of pixels of each N * N is had to identical position relationship with respect to described reference point is assigned in same group of hr pixel class, total N2-1 group hr pixel class in described hr image, generates N2-1 group hr pixel class image; Wherein, preferably first of described generation group of lr pixel class image and generate N2-1 group hr pixel class image all be according to identical resolution, according to the identical generation that puts in order (being that in different pixels class image, the pixel in same coordinate belongs to the some block of pixels in the block of pixels of above-mentioned N * N).

Each pixel in described first group of lr pixel class image and N2-1 group hr pixel class image all has respectively coordinate and the coordinate in hr image in the pixel class image under separately.

When the enlargement factor of described interpolation algorithm is N, preferably, described predetermined division rule is specially: each pixel in first group of lr pixel class of take is reference point or starting point, also by N * N, as large young pathbreaker hr image, mark off each block of pixels in the direction of the clock, and N2-1 group hr pixel class image (due to the relation of lr pixel class and lr image, can directly adopt lr image as lr pixel class image) is divided and generated to each block of pixels according to clockwise spiral order from outside to inside successively by above-mentioned N2-1 pixel except reference point.

When the enlargement factor of described interpolation algorithm is N, optionally, described predetermined division rule is: each pixel in first group of lr pixel class of take is reference point or starting point, by counterclockwise also marking off each block of pixels by N * N as large young pathbreaker hr image, and N2-1 group hr pixel class image is divided and generated to each block of pixels according to counterclockwise spiral order from outside to inside successively by above-mentioned N2-1 pixel except reference point.Described predetermined division rule in reality is not only confined to aforesaid way, can also be that other are according to interpolation amplification multiple N, according to N * N, as large young pathbreaker hr image, mark off block of pixels, as: spiral order or from inside to outside spiral order from outside to inside can be adopted, as shown in Figure 7.

The present embodiment is not limited to two kinds of modes above, " take first group of lr pixel class in each pixel be reference point " wherein described is to have identical position relationship because belong to the pixel of respectively organizing hr pixel class each block of pixels marking off from hr with the pixel of described lr pixel class, can also be specifically with other, there are identical characteristics point as a reference, describe and choosing of reference point do not done to particular determination herein.

The pixel at S102, the described edge of foundation determines to be the pixel at edge and the pixel that belongs to hr pixel class, respectively the pixel of each hr pixel class and the similarity distance of the pixel in described lr pixel class of calculative determination.

The computing method of similarity distance of the present invention are preferably: for calculating pixel (X1 in hr image, Y1) and pixel (Xk, the method of similarity distance Yk) is centered by described two pixels, to get respectively the first block of pixels and the second block of pixels first hr image to be divided into first group of lr pixel class image and the 2nd to N2 group hr pixel class image according to enlargement factor N, judge described (X1, Y1) and (Xk, Yk) two pixels belong to respectively first group of lr pixel class image and k group hr pixel class image, k ∈ [2, N2]; All pixels that belong to first group of lr pixel class obtain described the first block of pixels in first group of lr pixel class image in, and be classified as the 3rd block of pixels;

Describedly obtain all pixels that belong to first group of lr pixel class in described the first block of pixels and be specially: because each pixel in hr pixel class image has respectively coordinate and the coordinate in hr image in the pixel class image under separately, can get all pixels that belong to first group of lr pixel class in described the first block of pixels according to described corresponding relation;

All pixels that belong to k group hr pixel class obtain described the second block of pixels in k group hr pixel class image in, and be classified as the 4th block of pixels;

All pixels that belong to first group of lr pixel class obtain described the second block of pixels in first group of lr pixel class image in, and be classified as the 5th block of pixels;

All pixels that belong to k group hr pixel class obtain described the first block of pixels in k group hr pixel class image in, and be classified as the 6th block of pixels;

Calculate the 3rd block of pixels and the 4th block of pixels absolute error and obtain the first absolute error and, and calculate the 5th block of pixels and the 6th block of pixels absolute error and obtain the second absolute error and, by the first absolute error and with the second absolute error and addition, obtaining coordinate in first group of lr pixel class image is (x1, y1) in pixel and k group hr pixel class image, coordinate is the similarity distance between (xk, yk) pixel.Wherein the title of the 3rd block of pixels, the 4th block of pixels, the 5th block of pixels and the 6th block of pixels is only for convenient difference each other.

S103, obtain the pixel that similarity distance in lr pixel class meets similarity;

Preferably, in lr pixel class, each pixel is the pixel that meets similarity to P pixel of the similarity distance minimum of certain pixel of described definite hr pixel class.

Described P preferably comes to determine based on experience value, as: preferred, enlargement factor be 2 o'clock described in P to be preferably worth be 5;

Optionally, in enlargement factor, be not 2 o'clock, can also calculate acquisition by setting up funtcional relationship with enlargement factor, as P=2* enlargement factor+1.

Pixel in the lr pixel class that pixel in S104, described each the definite hr pixel class of foundation and correspondence meet similarity is constructed the parameter of non local nl wave filter;

Pixel in the lr pixel class that preferably pixel in described each the definite hr pixel class of foundation and correspondence meet similarity is constructed the parameter of non local nl wave filter, specifically comprises:

According to first pixel in described definite hr pixel class and corresponding described first pixel, meet pixel in P lr pixel class of similarity as parameter, calculate the weights of the pixel of described P lr pixel class;

The weights of the pixel of the coordinate of the pixel correspondence of preserving the coordinate of described first pixel correspondence in hr image, a described P lr pixel class in hr image and described P lr pixel class;

According to the method for first pixel in described hr pixel class, handle remaining pixel in described each definite hr pixel class, obtain determining to be coordinate in hr image of the pixel correspondence of each self-corresponding P lr pixel class that meets similarity of the pixel at edge and the pixel that belongs to hr pixel class and weights accordingly according to the pixel at described edge, and according to the coordinate of pixel correspondence in hr image in described each definite hr pixel class, meet the corresponding coordinate in hr image of pixel of the lr pixel class of similarity with pixel in described each definite hr pixel class, the weights that also have the pixel of corresponding described lr pixel class, corresponding relation between three is preserved, so just completed the parametric configuration of described nl wave filter.

Thereby S105, utilization have configured the nl wave filter filtration of described nl filter parameter and have obtained final enlarged image through the initial enlarged image of interpolator arithmetic.

Preferably, thus describedly utilize described nl wave filter to filter through the initial enlarged image of interpolator arithmetic to obtain final enlarged image, be specially:

The weights of the pixel of the coordinate of the pixel correspondence of obtaining the coordinate of described first pixel of preserving in the parameter of nl wave filter, a described P lr pixel class in hr image and described P lr pixel class;

The coordinate of pixel correspondence based on described P lr pixel class in hr image obtains P pixel value corresponding in initial enlarged image, and the weights of the pixel of described P lr pixel class of foundation obtain weighing computation results;

Described weighing computation results is replaced to the pixel value of corresponding described first pixel in initial enlarged image;

According to described first pixel method, handle successively the pixel of other hr pixel classes of preserving in the parameter of described nl wave filter, so just completed the nl filtering to described initial enlarged image.

The present embodiment 100 is by the hr image that interpolation algorithm is generated, according to pixel characteristic, be divided into lr pixel class and hr pixel class, and by avoid calculating in existing algorithm absolute error between the pixel of hr pixel class that there is no critical impact to finally solving similarity distance and calculating, on the basis that guarantees counting accuracy, improved the efficiency of existing algorithm.

The following examples 200 are the detailed descriptions to the lr pixel class in the present invention and the division of hr pixel class, and the hr image that the interpolation image algorithm that the enlargement factor of wherein take is 2 times generates is operand, as shown in Figure 2, specifically comprises:

S201, each pixel in first group of lr pixel class of take are reference point, in the direction of the clock hr image are divided into a plurality of 2 * 2 block of pixels, as shown in Figure 7.Wherein, by calculate the pixel coming from lr image in the hr image generating through interpolation algorithm, be classified as first group of lr pixel class, and generate first group of lr pixel class image.Each pixel that each 2 * 2 block of pixels be take in pixel class is reference point, by clockwise spiral order, the pixel except reference point is once classified as to pixel class 2, pixel class 3 and pixel class 4.

Allly in Fig. 7 indicate numeral 1 pixels to be referred to as pixel class 1 be lr pixel class, the pixel value of the pixel on its correspondence position is identical with the pixel value of the pixel of correspondence position in lr image, and all pixels that indicate numeral 2, numeral 3 and numeral 4 are called pixel class 2, pixel class 3 and pixel class 4, they all belong to hr pixel class, and the pixel value of its pixel is that the interpolation algorithm by pixel in lr image calculates.

For failing to be divided into according to described regulation the residual pixel point of 2 * 2 block of pixels, optional, the position according to this residual pixel point in predetermined division rule is divided into pixel class 1, pixel class 2, pixel class 3 or pixel class 4; Optionally, can also do to omit and process, because described residual pixel point is all the point on the frame in a little images, and nl filtration treatment is mainly in order to increase image border effect.Cause for failing to be divided into according to described regulation the residual pixel point of 2 * 2 block of pixels, when this situation may be division rule regulation, on framing mask pixel not can completely be divided into 2 * 2 block of pixels and cause.

The pixel of S202, extraction pixel class 1 generates first group of lr pixel class image L1, the pixel that extracts pixel class 2 generates second group of hr pixel class image L2, the pixel that extracts pixel class 3 generates the 3rd group of hr pixel class image L3, and the pixel that extracts pixel class 4 generates the 4th group of hr pixel class image L4; Lr pixel class image after generation and hr pixel class image are as shown in Figure 9.The mode that wherein generates each pixel class image is, the block of pixels that in every group of pixel class, pixel belongs to according to it puts in order according to 49 plain pieces shown in Fig. 7, and from left to right, order from top to bottom generates, and 4 pixel class images after generation as shown in Figure 9.

Wherein, pixel class image L1, pixel class image L2, pixel class image L3 are identical with pixel class image L4 resolution sizes, and the pixel in each pixel class image has coordinate and the respective coordinates in hr image in each pixel class image under comfortable.

S203, the coordinate mapping relations of preserving each pixel coordinate and corresponding pixel points in hr image in first group of lr pixel class image, second group of hr pixel class image, the 3rd group of hr pixel class image and the 4th group of hr pixel class image.

Preferably, the coordinate of the true origin of preserving each pixel class image in hr image, thus in conjunction with division rule and enlargement factor, just can complete the conversion of same pixel point coordinate in pixel coordinate and hr image in each pixel class image.

For example: in first group of lr pixel class image, coordinate is (1,1) pixel, being converted to pixel coordinate in hr image is specially: first according to as a reference point with pixel class 1 each pixel, and clockwise division rule draws, the initial point of pixel class 1 image and the origin offset off_origina1_x=off_origina1_y=0 of hr image, and enlargement factor is 2; In first group of lr pixel class image, coordinate is that the coordinate of the pixel of (1,1) corresponding identical pixel in hr image is (off_origina1_x+1*2, off_origina1_y+1*2)=(2,2).

Optionally, can also and respectively organize pixel class image for distinguishing by hr image, the corresponding pixel correspondence respectively organized in the pixel class image coordinate in hr image of preserving, its preservation form can be as shown in the table:

Wherein symbol " " there is not corresponding coordinate points in representative, (X1 wherein, Y1) according to the pixel number at the edge in hr image as footnote, and (x21, y21) be using successively pixel class image that this edge hr pixel belonged to be which group and in table in corresponding hr pixel class image column appearance order as footnote.

By embodiment 200, completed the generation of each pixel class image, the generation of this pixel class image is according to interpolation algorithm characteristic, and the pixel that has a hr pixel class of identical offset relationship with the pixel of lr pixel class in hr image is planned in same group of hr pixel class image.Although this embodiment realizes based on enlargement factor 2, the technology contents that those skilled in the art can disclose according to the present embodiment, derives the method in other enlargement factors and uses, and does not repeat them here.Embodiment 300 then will calculate existing similarity distance to do further simplification on this basis.

A kind of enlarged image embodiment of the method 300 provided by the invention, the present embodiment is based on embodiment of the method 200, for the concrete utilization of embodiment of the method 100.The present embodiment relates to the block of pixels not relating in embodiment 200, block of pixels in embodiment 200 is to use for dividing pixel class, and in the present embodiment, relate to the block of pixels for the similarity distance between calculating pixel point, the essential distinction of block of pixels be between the two the rule of its division and purposes not identical.Enlargement factor in the present embodiment is 2 times, and the building method of the existing parameter with the non local nl wave filter in interpolator arithmetic illustrates the improvement of the present embodiment in speed, and as shown in Figure 3, its process is specially its process flow diagram:

As shown in Figure 8, the pixel and the coordinate that for coordinates computed in prior art, are B_L1 ((cx-1) * 2+1, (cy-1) * 2+1) are the similarity schematic diagram of the pixel of B_L2 ((x-1) * 2, (y-1) * 2+1).First construct ((cx-1) * 2+1 with coordinate B_L1, (cy-1) the first block of pixels B_L1 that 7 pixels of take * 2+1) form as the length of side, then structure is with coordinate B_L2 ((x-1) * 2, (y-1) the second block of pixels B_L2 that 7 pixels of take * 2+1) form as the length of side, then by correspondence position pixel in block of pixels B_L1 and block of pixels B_L2, 49 pixels ask respectively absolute error and, described absolute error and value to be coordinate be B_L1 ((cx-1) * 2+1, (cy-1) * 2+1) pixel and coordinate are B_L2 ((x-1) * 2, (y-1) * 2+1) pixel between similarity distance.

The present invention finds by analyzing the computation process of similarity distance in above-mentioned prior art, in the process of the similarity distance of calculating pixel point 1 and pixel 2 real effectively actual be matrix that in B_L1, pixel 1 and pixel 2 form with B_L2 in the matrix that forms of pixel 2 and pixel 1 absolute error with; Pixel 3 and pixel 4 are to be calculated and obtained by interpolation algorithm (can also adopt other interpolator arithmetics in enlarged image), the absolute error of pixel 3 and pixel 4 and computation process in the situation that having increased overall calculation amount, but can not bring the raising of the accuracy of similarity distance), according to described analysis, further obtain the inventive method.

S301, based on each reference point, mark off in the direction of the clock each 2 * 2 block of pixels, described reference point is in the present embodiment for originally belonging to the pixel of lr image; In each 2 * 2 block of pixels that form based on described reference point, recording each reference point is first group of lr pixel class, and its excess-three hr pixel of described each 2 * 2 block of pixels of record is second group of hr pixel class, the 3rd group of hr pixel class and the 4th group of hr pixel class in the direction of the clock;

S302, the pixel comprising for described first group of lr pixel class, second group of hr pixel class, the 3rd group of hr pixel class and the 4th group of hr pixel class respectively generate corresponding first group of lr pixel class image L1, second group of hr pixel class image L2, the 3rd group of hr pixel class image L3 and the 4th group of hr pixel class image L4, as shown in Figure 9.The mapping relations of each pixel in each pixel and hr image in preservation pixel class image L1, L2, L3 and L4.

S303, according to the pixel at edge, determine in hr image to be the pixel at edge and the pixel that belongs to hr pixel class, the respectively pixel of each hr pixel class and the similarity distance of the pixel in described lr pixel class of calculative determination, and obtain the pixel that similarity distance in lr pixel class meets similarity;

When solving the pixel coordinate at edge, be that B_L2 ((x-1) * 2, (y-1) * 2+1) and the coordinate that belongs to lr pixel class are while being B_L1 ((cx-1) * 2+1, (cy-1) * 2+1) pixel pixel similarity distance;

It (is that all pixels 1 and all pixels in B_L2 block of pixels 2 in B_L1 block of pixels have carried out absolute error and computing for the first time that B_L1 block of pixels (being the first block of pixels in S102) and B_L2 block of pixels (being the second block of pixels in S102) have all comprised pixel 1 and pixel 2, all pixels 2 and all pixels in B_L2 block of pixels 1 in B_L1 block of pixels have carried out absolute error and computing for the second time), and the first group of lr pixel class image L1 and the second group of hr pixel class image L2 that generate have all only comprised of a sort pixel 1 and pixel 2, therefore complete L1 (cx, xy) block of pixels centered by (being the 3rd block of pixels in S102) and L2 (x, the absolute error of the block of pixels y) (being the 4th block of pixels in S102) and calculate after the absolute error of all pixels 2 in all pixels 1 to the B_L2 block of pixels in B_L1 block of pixels (be and), also to calculate L1 (x, y) block of pixels centered by (being the 5th block of pixels in S102) and L2 (cx-1, the absolute error of the block of pixels cy) (being the 6th block of pixels in S102) and the absolute error of all pixels 2 in all pixels 1 to the B_L1 block of pixels in B_L2 block of pixels (be and), and will after 2 absolute errors obtained above and results added, get average, the result of getting average after described addition be exactly in Fig. 8 coordinate be B_L1 ((cx-1) * 2+1, (cy-1) * 2+1) pixel and coordinate are B_L2 ((x-1) * 2, (y-1) * 2+1) pixel between similarity distance, L1 (cx in above-mentioned pixel class image 1, xy) and L1 (x, y) L2 (x and in pixel class image 2, y) and L2 (cx-1, cy) confirmation of coordinate, can the mapping relations to first group of lr pixel class image L1 and second group of hr pixel class image L2 be obtained by B_L1 block of pixels and B_L2 block of pixels, described mapping relations can adopt in S203 any one.

The parameter of S304, structure nl wave filter.

The parametric technique of structure nl wave filter specifically comprises:

According to B_L2 ((x-1) * 2 in described definite hr pixel class, (y-1) * 2+1) pixel and corresponding described B_L2 ((x-1) * 2, (y-1) * 2+1) pixel meets pixel in P lr pixel class of similarity as parameter, calculates the weights of the pixel of described P lr pixel class;

The weights of the pixel of the coordinate of the pixel correspondence of preserving the coordinate of described B_L2 ((x-1) * 2, (y-1) * 2+1) pixel correspondence in hr image, a described P lr pixel class in hr image and described P lr pixel class;

Wherein the method for B_L2 ((x-1) * 2, (y-1) * 2+1) pixel calculating weights is specially:

Corresponding B_L2 ((x-1) * 2, (y-1) * 2+1) pixel record is P pixel coordinate that meets the L1 of similarity of correspondence with it, and according to formula i=1 wherein, 2, ..., P calculates the weights of P lr pixel, wherein, SAD (i) is B_L2 ((x-1) * 2, (y-1) * 2+1) pixel and i similarity distance that meets the pixel of similarity, t is the parameter of analogue noise attenuation degree, and n is the size of block of pixels in hr image.

Method according to B_L2 ((x-1) * 2, (y-1) * 2+1) pixel in described hr pixel class, has traveled through remaining pixel in described each definite hr pixel class according to S303.Just completed the structure of the parameter of nl wave filter, the parameter of this wave filter can be that the form with Figure 11 exists; Wherein first row is in hr image, to be the pixel at edge and the pixel coordinate of hr pixel class, and secondary series is the coordinate of P lr pixel class point in hr image that meets similarity, and the 3rd class is the weights that meet P lr pixel class point of similarity.In the present embodiment, in enlargement factor, be 2 o'clock, it is 5 that described P is preferably worth.

A kind of enlarged image embodiment of the method 400 provided by the invention, the application of the parameter embodiment that the present embodiment is shown said n l wave filter by concrete grammar flow process in enlarged image algorithm, and after extracting the pixel at edge, carried out edge-diffusion processing, the filter effect of nl wave filter is further enhanced.The present embodiment adopts 2 times to amplify interpolation algorithm as description object, and as shown in Figure 4, its flow process is specially:

S401, low resolution lr image is generated to high resolving power hr image by image interpolation algorithm.

Described image interpolation algorithm can be point of proximity interpolation algorithm, bilinear interpolation algorithm, two cubes of interpolation algorithms or self-adaptation Based on Interpolating Spline.

S402, described hr image is carried out to the pixel that rim detection is obtained edge in image.

Described edge detection method can be but be not limited to any one in Roberts, Sobel, Prewitt, LOG or Canny operator.

S403, the edge contour consisting of the pixel at edge in image is carried out to edge-diffusion processing.

In concrete enforcement, edge-diffusion can adopt various ways, such as usining the pixel at each edge as central point, according to specific shape, spread, all points in this shape all belong to the edge after diffusion, and described specific shape can be rhombus, square etc.The size of described specific shape is preferably got the minimum odd number that is more than or equal to enlargement factor, as while adopting square and enlargement factor to be 2 times, this foursquare length of side is 3 pixels; As while adopting square and enlargement factor to be 3 times, this foursquare length of side is 3 pixels; As while adopting square and enlargement factor to be 4 times, this foursquare length of side is 5.

S404, by the parameter of the step S301-S304 structure nl wave filter in embodiment of the method 300.

In the present invention except adopting the mapping relations of S303 to obtain respective coordinates relation, can also adopt the model calculating according to pixel and other similarity distances of respectively organizing the pixel of hr pixel class with first group of lr pixel class, be used as follow-up computing method, (x1, y1) be coordinate in first group of lr pixel class, (xk, yk) is coordinate in k group hr pixel class; Contrast Fig. 8 and its model acquisition methods of Fig. 9 are:

According to the similarity distance of the first block of pixels centered by the second block of pixels centered by (X1, Y1) in hr image and (Xk, Yk), wherein (X1, Y1) and (Xk, Yk) is respectively coordinate (x1, y1) and (xk, yk) respective coordinates in hr image;

All pixels that belong to first group of lr pixel class obtain described the first block of pixels (described in Fig. 8, the first block of pixels shows as B_L1) in first group of lr pixel class image in, and be classified as the 3rd block of pixels; All pixels that belong to k group hr pixel class obtain described the second block of pixels in k group hr pixel class image in, and be classified as the 4th block of pixels; All pixels that belong to first group of lr pixel class obtain described the second block of pixels in first group of lr pixel class image in, and be classified as the 5th block of pixels; All pixels that belong to k group hr pixel class obtain described the first block of pixels in k group hr pixel class image in, and be classified as the 6th block of pixels

Calculate the 3rd block of pixels and the 4th block of pixels absolute error and, and calculate the 5th block of pixels and the 6th block of pixels absolute error and, then two absolute errors just can be obtained to coordinate in first group of lr pixel class with summation is (x1, y1) with Group X hr pixel class in coordinate be the similarity distance between (xX, yX).According to said process, the 3rd block of pixels, the 4th block of pixels, the 5th block of pixels and the 6th block of pixels are generated to corresponding model.

The enlargement factor of the present embodiment is 2, for the 3rd block of pixels, the 4th block of pixels, the 5th block of pixels and the 6th block of pixels of above-mentioned formation, generates corresponding model, specifically describes as follows:

Pixel according to described edge obtains pixel corresponding in pixel class image L1, pixel class image L2, pixel class image L3 or pixel class image L4, and recording this pixel coordinate getting is (x, y); Travel through pixel in first group of lr pixel class, and calculate with described in the similarity distance of pixel of (x, the y) that traverse and first group of lr pixel class, its process is specially:

When (x, y) pixel is in L2, travel through successively the similarity of each pixel and (x, y) in L1.Suppose the current L1 of traversing (cx, cy) time, to calculate in L1 with point (cx, cy) in the 3rd block of pixels model centered by and L2 with point (x, the SAD of the 4th block of pixels model y) adds in L1 with point (x, the SAD of the 6th block of pixels model in the 5th block of pixels model y) and L2 centered by point (cx-1, cy), get two SAD's and as final sad value.

When (x, y) pixel is in L3, travel through successively the similarity of each pixel and (x, y) in L1.Suppose the current L1 of traversing (cx, cy) time, to calculate in L1 with point (cx, cy) in the 3rd block of pixels model centered by and L3 with point (x, the SAD of the 4th block of pixels model y) adds in L1 with point (x, the SAD of the 6th block of pixels model in the 5th block of pixels model y) and L3 centered by point (cx, cy-1), get two SAD's and as final sad value.

When (x, y) pixel is in L4, travel through successively the similarity of each pixel and (x, y) in L1.Suppose the current L1 of traversing (cx, cy) time, to calculate in L1 with point (cx, cy) in the 3rd block of pixels model centered by and L4 with point (x, the SAD of the 4th block of pixels model y) adds in L1 with point (x, the SAD of the 6th block of pixels model in the 5th block of pixels model y) and L4 centered by point (cx--1, cy-1), get two SAD's and as final sad value.

When described the 3rd block of pixels and the 4th block of pixels size are all M * M, the absolute error of described calculating the 3rd block of pixels and the 4th block of pixels and, specifically comprise: the pixel value that obtains each pixel in the 3rd block of pixels is v1, v2, v3, ..., in vn (M * M=n) and the 4th block of pixels, the pixel value of corresponding pixel points is u1, u2, u3, ..., un; And by formula S AD=|u1-v1|+|u2-v2|+|u3-v3|+......+|un-vn| calculate the 3rd block of pixels and the 4th block of pixels absolute error and.

In lr pixel class, each pixel is the pixel that meets similarity to P pixel of the similarity distance minimum of certain pixel of hr pixel class.When enlargement factor is 2 times, described P value preferably 5.

Record lr pixel correspondence that each hr pixel and corresponding P the meet similarity coordinate in hr image, calculate the weights of P lr pixel.

S405, utilize interpolator arithmetic to amplify lr image.

Described interpolator arithmetic can be point of proximity interpolation algorithm, bilinear interpolation algorithm, two cubes of interpolation algorithms or self-adaptation Based on Interpolating Spline.

S406, utilize nl wave filter to filter described image after interpolator arithmetic amplifies.

The weights of the pixel of the coordinate of the pixel correspondence of obtaining the coordinate of described first pixel of preserving in the parameter of nl wave filter, a described P lr pixel class in hr image and described P lr pixel class; The coordinate of pixel correspondence based on described P lr pixel class in hr image obtains P pixel value corresponding in initial enlarged image, and the weights of the pixel of described P lr pixel class of foundation obtain weighing computation results; Described weighing computation results is replaced to the pixel value of corresponding described first pixel in initial enlarged image; Pixel according to other hr pixel classes of preserving in described first pixel method successively parameter based on described nl wave filter is processed initial enlarged image.

Wherein also the weights of the pixel of described P lr pixel class of foundation obtain weighing computation results, and concrete grammar is:

The big or small n that obtains parametric t and the block of pixels of analogue noise attenuation degree, show in enlarged image embodiment of the method 300, the preferred value of n is 7 when enlargement factor is 2 times; According to formula i=1,2 ..., P calculates the weights for the pixel of P lr pixel class; Wherein, SAD (i) be i point and the pixel of corresponding hr pixel class in the pixel of P lr pixel class absolute error and;

Obtain P lr pixel the pixel value H (i) of pixel, i=1,2 ..., P, and utilize formula calculate pixel value and replace the pixel value of hr pixel, wherein represent that current high-resolution position pixel participates in the weights that weighted sum is used, it is the maximal value of getting in P weights above;

Wherein ω (i) is the weights of the pixel of P lr pixel class, and the big or small n of block of pixels in the parametric t by analogue noise attenuation degree and hr image, according to formula i=1 wherein, 2 ..., P calculates acquisition.

The present embodiment is by conjunction with concrete computational algorithm, and how the parameter of the wave filter that embodiment of the method 300 is before obtained specifically makes to be used as refinement description, utilize this method embodiment just can completely realize the enlarged image method that this method will realize.By avoid calculating in existing algorithm absolute error between the pixel of hr pixel class that there is no critical impact to finally solving similarity distance and calculating, on the basis that guarantees counting accuracy, improved the efficiency of existing algorithm.

A kind of enlarged image device embodiment 500 provided by the invention, its structural representation as shown in Figure 5, specifically comprises:

Pixel sort module 101, for by the pixel of described hr image according to pixels characteristic be divided into lr pixel class and hr pixel class, described lr pixel class comprises the pixel coming from lr image, described hr pixel class comprises the pixel calculate generating through interpolation algorithm;

Similarity distance computing module 102, for determining to be the pixel at edge and the pixel that belongs to hr pixel class according to described edge image, the pixel of each hr pixel class and the similarity distance of the pixel in described lr pixel class of difference calculative determination;

Similitude acquisition module 103, meets the pixel of similarity for obtaining lr pixel class similarity distance;

The parameter generation module 104 of non local nl wave filter, for constructing the parameter of non local nl wave filter according to the pixel in the pixel of described each definite hr pixel class and the corresponding lr pixel class that meets similarity;

Filtering module 105, thus for utilizing the parameter of described nl wave filter to filter through the initial enlarged image of interpolator arithmetic, obtain final enlarged image.

Preferably, described pixel sort module 101, also comprises:

Pixel class image generates submodule 201, is classified as first group of lr pixel class, and generates first group of lr pixel class image for the pixel that the hr image that calculates generation through interpolation algorithm is come from lr image;

Each pixel in first group of lr pixel class of take is reference point, by predetermined division rule, hr image is divided into the block of pixels of a plurality of N * N, the enlargement factor that wherein N is interpolation algorithm, and N is greater than 1 natural number; And the pixel that the block of pixels of each N * N is had to identical position relationship with respect to described reference point assigns in same group of hr pixel class, total N2-1 group hr pixel class in described hr image, and generate N2-1 group hr pixel class image; Each pixel in described first group of lr pixel class image and N2-1 group hr pixel class image all has respectively coordinate and the coordinate in hr image in pixel class image separately.

Preferably, described similarity distance computing module 102 also comprises mapping generation module 202, be used for obtaining hr image with (X1, Y1) the first block of pixels centered by and (Xk, Yk) the second block of pixels centered by, wherein (X1, Y1) is that in first group of lr pixel class image, coordinate is the pixel correspondence of (x1, the y1) coordinate in hr image; Wherein (Xk, Yk) is pixel correspondence that in k group hr pixel class image, coordinate is (xk, the yk) coordinate in hr image;

All pixels that belong to first group of lr pixel class obtain described the first block of pixels in first group of lr pixel class image in, and be classified as the 3rd block of pixels;

All pixels that belong to k group hr pixel class obtain described the second block of pixels in k group hr pixel class image in, and be classified as the 4th block of pixels;

All pixels that belong to first group of lr pixel class obtain described the second block of pixels in first group of lr pixel class image in, and be classified as the 5th block of pixels;

All pixels that belong to k group hr pixel class obtain described the first block of pixels in k group hr pixel class image in, and be classified as the 6th block of pixels.

Now, described similarity distance computing module 102, also for calculate the 3rd block of pixels and the 4th block of pixels absolute error and obtain the first absolute error and, and calculate the 5th block of pixels and the 6th block of pixels absolute error and obtain the second absolute error and, by the first absolute error and with the second absolute error and addition, in the pixel that to obtain coordinate in first group of lr pixel class image be (x1, y1) and k group hr pixel class image, coordinate is the similarity distance between (xk, yk) pixel.

A kind of enlarged image device embodiment 600 provided by the invention, as shown in Figure 6.Comprise at least one processor (301), CPU for example, at least one network interface 304 or other communication interfaces 303, storer 305, and at least one communication bus 302.Communication bus 302 is for realizing the connection communication between these devices.Optional user interface 903 can be display, keyboard or pointing device.Storer 305 may comprise high speed Ram storer, also may also comprise non-unsettled storer (non-volatilememory), for example at least one magnetic disk memory.Storer 305 optionally can comprise at least one and be positioned at the memory storage away from aforementioned CPU302.In some embodiments, storer 305 has been stored following element, module or data structure, or their subset, or their superset:

Operating system 306, comprises various programs, for realizing various basic businesses and processing hardware based task;

Application module 307, comprise as lower module combination: pixel sort module 101, similarity distance computing module 102, similitude acquisition module 103, the parameter generation module 104 of non local nl wave filter, filtering module 105, the function of above-mentioned module can be with reference to the declaratives of the fundamental diagram of figure 3 or Fig. 4, also can be with reference to the declaratives of figure 5; Repeat no more herein.

Those of ordinary skills can recognize, in conjunction with the various method steps of describing in embodiment disclosed herein and unit, can realize with electronic hardware, computer software or the combination of the two, for the interchangeability of hardware and software is clearly described, step and the composition of each embodiment described according to function in the above description in general manner.These functions are carried out with hardware or software mode actually, depend on application-specific and the design constraint of technical scheme.Those of ordinary skills can specifically should be used for realizing described function with distinct methods to each, but this realization should not thought and exceeds scope of the present invention.

The software program that the method for describing in conjunction with embodiment disclosed herein or step can use hardware, processor to carry out, or the combination of the two is implemented.Software program can be placed in the storage medium of any other form known in random access memory (RAM), internal memory, ROM (read-only memory) (ROM), electrically programmable ROM, electrically erasable ROM, register, hard disk, moveable magnetic disc, CD-ROM or technical field.

Although the present invention be have been described in detail by reference to accompanying drawing mode in conjunction with the preferred embodiments, the present invention is not limited to this.Without departing from the spirit and substance of the premise in the present invention, those of ordinary skills can carry out to embodiments of the invention modification or the replacement of various equivalences, and these modifications or replacement all should be in covering scopes of the present invention.

Claims (13)

1. a method for enlarged image, generates high resolving power (hr) image by low resolution (lr) image by image interpolation algorithm, and described hr image is carried out to the pixel that rim detection is obtained edge in image, it is characterized in that, comprising:
By the pixel of described hr image according to pixels characteristic be divided into lr pixel class and hr pixel class, described lr pixel class comprises the pixel coming from lr image, described hr pixel class comprises the pixel calculate generating through interpolation algorithm;
According to the pixel at described edge, determine to be the pixel at edge and the pixel that belongs to hr pixel class, calculate respectively the pixel of described each definite hr pixel class and the similarity distance of the pixel in described lr pixel class, and obtain the pixel that similarity distance in lr pixel class meets similarity;
According to the pixel in the pixel in described each definite hr pixel class and the corresponding lr pixel class that meets similarity, construct the parameter of non local nl wave filter;
Thereby utilize described nl wave filter to filter through the initial enlarged image of interpolator arithmetic and obtain final enlarged image.
2. method according to claim 1, is characterized in that, when the enlargement factor of described interpolation algorithm is N, wherein, N is greater than 1 natural number, described by described hr image according to pixels characteristic be divided into lr pixel class and hr pixel class, specifically comprise:
By calculate the pixel coming from lr image in the hr image generating through interpolation algorithm, be classified as first group of lr pixel class, and generate first group of lr pixel class image;
Each pixel in first group of lr pixel class of take is reference point, by predetermined division rule, hr image is divided into the block of pixels of a plurality of N * N, and the pixel that the block of pixels of each N * N is had to identical position relationship with respect to described reference point assigns in same group of hr pixel class, total N in described hr image 2-1 group of hr pixel class, generates N 2-1 group of hr pixel class image;
Described first group of lr pixel class image and N 2each pixel in-1 group of hr pixel class image has respectively coordinate and the coordinate in hr image in the pixel class image under separately.
3. method according to claim 2, it is characterized in that, it is the pixel at edge and the pixel that belongs to hr pixel class that the pixel at the described edge of described foundation is determined, wherein, when specifically determining pixel that coordinate is the edge of (X, Y), meet while belonging to hr pixel class, calculate the pixel (X of described definite hr pixel class, the process of the similarity distance of the pixel Y) and in described lr pixel class, specifically comprises:
Pixel coordinate (X according to described edge, Y), in judgement (X, Y), belong to after the pixel in hr pixel class, determine that described pixel specifically belongs to the hr pixel class image of k group, and the coordinate in the hr pixel class image of described k group is (x k, y k), k ∈ [2, N wherein 2];
Calculating respectively coordinate in the hr pixel class image of each pixel and described k group in first group of lr pixel class image is (x k, y k) the similarity distance of pixel.
4. method according to claim 3, is characterized in that, in specifically calculating first group of lr pixel class image, coordinate is (x 1, y 1) coordinate is (x in pixel and k group hr pixel class image k, y k) the similarity distance of pixel time, be specially:
Obtain in hr image with (X 1, Y 1) centered by the first block of pixels and (X k, Y k) centered by the second block of pixels, (X wherein 1, Y 1) be that in first group of lr pixel class image, coordinate is (x 1, y 1) the coordinate of pixel correspondence in hr image; (X wherein k, Y k) be that in k group hr pixel class image, coordinate is (x k, y k) the coordinate of pixel correspondence in hr image;
All pixels that belong to first group of lr pixel class obtain described the first block of pixels in first group of lr pixel class image in, and be classified as the 3rd block of pixels;
All pixels that belong to k group hr pixel class obtain described the second block of pixels in k group hr pixel class image in, and be classified as the 4th block of pixels;
All pixels that belong to first group of lr pixel class obtain described the second block of pixels in first group of lr pixel class image in, and be classified as the 5th block of pixels;
All pixels that belong to k group hr pixel class obtain described the first block of pixels in k group hr pixel class image in, and be classified as the 6th block of pixels;
Calculate the 3rd block of pixels and the 4th block of pixels absolute error and obtain the first absolute error and, and calculate the 5th block of pixels and the 6th block of pixels absolute error and obtain the second absolute error and, by the first absolute error and with the second absolute error and addition, obtaining coordinate in first group of lr pixel class image is (x 1, y 1) pixel and k group hr pixel class image in coordinate be (x k, y k) similarity distance between pixel.
5. according to the arbitrary described method of claim 2-4, it is characterized in that, at described enlargement factor N, it is 2 o'clock, it is described that to take each pixel in first group of lr pixel class be reference point, by predetermined division rule, hr image is divided into the block of pixels of a plurality of N * N, and the pixel that the block of pixels of each N * N is had to identical position relationship with respect to described reference point assigns in same group of hr pixel class, total N in described hr image 2-1 group of hr pixel class, and generate N 2-1 group of hr pixel class image, specifically comprises:
Each pixel in first group of lr pixel class of take is reference point, in the direction of the clock described hr image is divided into each block of pixels of 2 * 2;
In each 2 * 2 block of pixels of dividing out based on described reference point, so that described reference point is for referencial use, successively the pixel of its excess-three hr pixel class except described reference point in described each 2 * 2 block of pixels is belonged to respectively to second group of hr pixel class, the 3rd group of hr pixel class and the 4th group of hr pixel class in the direction of the clock;
Respectively according to described first group of lr pixel class, second group of hr pixel class, the 3rd group of hr pixel class and the 4th group of corresponding first group of lr pixel class image, second group of hr pixel class image, the 3rd group of hr pixel class image and the 4th group of hr pixel class image of pixel generation that hr pixel class comprises.
6. according to the method described in claim 4 or 5, it is characterized in that, described the 3rd block of pixels and the 4th block of pixels size are all M * M, the absolute error of described calculating the 3rd block of pixels and the 4th block of pixels and, specifically comprise:
The pixel value that obtains each pixel in the 3rd block of pixels is v 1, v 2, v 3..., in vn (M * M=n) and the 4th block of pixels, the pixel value of corresponding pixel points is u 1, u 2, u 3..., u n; And by formula S AD=|u 1-v 1|+| u 2-v 2|+| u 3-v 3|+...+| u n-v n| calculate the 3rd block of pixels and the 4th block of pixels absolute error and.
7. according to the arbitrary described method of claim 1-6, it is characterized in that, the pixel of each hr pixel class of the described calculative determination of described correspondence, obtains the pixel that similarity distance in lr pixel class meets similarity, specifically comprises:
The pixel of each hr pixel class of corresponding described calculative determination, obtains P pixel of similarity distance minimum in lr pixel class as the pixel that meets similarity, and wherein P is greater than 1 natural number.
8. according to the arbitrary described method of claim 2-7, it is characterized in that, the pixel in the pixel in described each the definite hr pixel class of described foundation and the corresponding lr pixel class that meets similarity is constructed the parameter of non local nl wave filter, specifically comprises:
According to first pixel in described definite hr pixel class and corresponding described first pixel, meet pixel in P lr pixel class of similarity as parameter, calculate the weights of the pixel of described P lr pixel class;
The weights of the pixel of the coordinate of the pixel correspondence of preserving the coordinate of described first pixel correspondence in hr image, a described P lr pixel class in hr image and described P lr pixel class;
According to the method for first pixel in described hr pixel class, handle remaining pixel in described each definite hr pixel class.
9. method according to claim 8, is characterized in that, thereby the described nl of utilization wave filter filters through the initial enlarged image of interpolator arithmetic and obtains final enlarged image, specifically comprises:
The weights of the pixel of the coordinate of the pixel correspondence of obtaining the coordinate of described first pixel of preserving in the parameter of nl wave filter, a described P lr pixel class in hr image and described P lr pixel class;
The coordinate of pixel correspondence based on described P lr pixel class in hr image obtains P pixel value corresponding in initial enlarged image, and the weights of the pixel of described P lr pixel class of foundation obtain weighing computation results;
Described weighing computation results is replaced to the pixel value of corresponding described first pixel in initial enlarged image;
Pixel according to other hr pixel classes of preserving in described first pixel method successively parameter based on described nl wave filter is processed initial enlarged image.
10. method according to claim 9, it is characterized in that, the coordinate of the described pixel correspondence based on described P lr pixel class in hr image obtains P pixel value corresponding in initial enlarged image, and obtain weighing computation results according to the weights of the pixel of described P lr pixel class, specifically comprise:
According to the P preserving in the parameter of nl wave filter and described first pixel, meet the coordinate of the pixel of similarity, obtain and amplify the pixel value H (i) on a described P coordinate in image afterwards, i=1,2 ..., P, and utilize formula calculate weighing computation results, wherein represent that current high-resolution position pixel participates in the weights that weighted sum is used, it is the maximal value of getting in P weights above;
Wherein ω (i) is the weights of the pixel of P lr pixel class, and the big or small n of block of pixels in the parametric t by analogue noise attenuation degree and hr image, according to formula i=1 wherein, 2 ..., P calculates acquisition.
11. 1 kinds of enlarged image devices, is characterized in that, comprising:
Pixel sort module, for by the pixel of described hr image according to pixels characteristic be divided into lr pixel class and hr pixel class, described lr pixel class comprises the pixel coming from lr image, described hr pixel class comprises the pixel calculate generating through interpolation algorithm;
Similarity distance computing module, determines to be the pixel at edge and the pixel that belongs to hr pixel class for the pixel according to described edge, respectively the pixel of each hr pixel class and the similarity distance of the pixel in described lr pixel class of calculative determination;
Similitude acquisition module, meets the pixel of similarity for obtaining lr pixel class similarity distance;
The parameter generation module of non local nl wave filter, for constructing the parameter of non local nl wave filter according to the pixel in the pixel of described each definite hr pixel class and the corresponding lr pixel class that meets similarity;
Filtering module, thus for utilizing nl wave filter to filter through the initial enlarged image of interpolator arithmetic, obtain final enlarged image.
12. devices according to claim 11, is characterized in that, described pixel sort module, also comprises:
Pixel class image generates submodule, is classified as first group of lr pixel class, and generates first group of lr pixel class image for the pixel that the hr image that calculates generation through interpolation algorithm is come from lr image;
Each pixel in first group of lr pixel class of take is reference point, by predetermined division rule, hr image is divided into the block of pixels of a plurality of N * N, and the pixel that the block of pixels of each N * N is had to identical position relationship with respect to described reference point assigns in same group of hr pixel class, total N in described hr image 2-1 group of hr pixel class, and generate N 2-1 group of hr pixel class image; Wherein, the enlargement factor that N is described interpolation algorithm, N is greater than 1 natural number;
Described first group of lr pixel class image and N 2each pixel in-1 group of hr pixel class image all has respectively coordinate and the coordinate in hr image in pixel class image separately.
13. devices according to claim 11, is characterized in that, described similarity distance computing module also comprises mapping generation module, for ought specifically calculating first group of lr pixel class image coordinate, are (x 1, y 1) coordinate is (x in pixel and k group hr pixel class image k, y k) the similarity distance of pixel time, be specially:
Obtain in hr image with (X 1, Y 1) centered by the first block of pixels and (X k, Y k) centered by the second block of pixels, (X wherein 1, Y 1) be that in first group of lr pixel class image, coordinate is (x 1, y 1) the coordinate of pixel correspondence in hr image; (X wherein k, Y k) be that in k group hr pixel class image, coordinate is (x k, y k) the coordinate of pixel correspondence in hr image;
All pixels that belong to first group of lr pixel class obtain described the first block of pixels in first group of lr pixel class image in, and be classified as the 3rd block of pixels;
All pixels that belong to k group hr pixel class obtain described the second block of pixels in k group hr pixel class image in, and be classified as the 4th block of pixels;
All pixels that belong to first group of lr pixel class obtain described the second block of pixels in first group of lr pixel class image in, and be classified as the 5th block of pixels;
All pixels that belong to k group hr pixel class obtain described the first block of pixels in k group hr pixel class image in, and be classified as the 6th block of pixels;
Described similarity distance computing module, also for calculate the 3rd block of pixels and the 4th block of pixels absolute error and obtain the first absolute error and, and calculate the 5th block of pixels and the 6th block of pixels absolute error and obtain the second absolute error and, by the first absolute error and with the second absolute error and addition, obtaining coordinate in first group of lr pixel class image is (x 1, y 1) pixel and k group hr pixel class image in coordinate be (x k, y k) similarity distance between pixel.
CN201210346575.XA 2012-09-18 2012-09-18 A kind of method of enlarged drawing CN103679631B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201210346575.XA CN103679631B (en) 2012-09-18 2012-09-18 A kind of method of enlarged drawing

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201210346575.XA CN103679631B (en) 2012-09-18 2012-09-18 A kind of method of enlarged drawing
PCT/CN2013/071098 WO2014044027A1 (en) 2012-09-18 2013-01-30 Method for enlarging image

Publications (2)

Publication Number Publication Date
CN103679631A true CN103679631A (en) 2014-03-26
CN103679631B CN103679631B (en) 2018-01-23

Family

ID=50317089

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201210346575.XA CN103679631B (en) 2012-09-18 2012-09-18 A kind of method of enlarged drawing

Country Status (2)

Country Link
CN (1) CN103679631B (en)
WO (1) WO2014044027A1 (en)

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101833753A (en) * 2010-04-30 2010-09-15 西安电子科技大学 SAR image de-speckling method based on improved Bayes non-local mean filter
US7856150B2 (en) * 2007-04-10 2010-12-21 Arcsoft, Inc. Denoise method on image pyramid
CN102298773A (en) * 2011-09-19 2011-12-28 西安电子科技大学 Shape-adaptive non-local mean denoising method
US20120189051A1 (en) * 2008-04-11 2012-07-26 Yunfei Zheng Methods and apparatus for deblocking filtering of non-local intra prediction

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
ES2382832T3 (en) * 2007-10-02 2012-06-13 Vestel Elektronik Sanayi Ve Ticaret A.S. Method and apparatus for changing the spatial resolution of a digital image
CN101499164B (en) * 2009-02-27 2011-02-09 西安交通大学 Image interpolation reconstruction method based on single low-resolution image
CN101661611B (en) * 2009-09-25 2011-08-24 西安电子科技大学 Realization method based on bayesian non-local mean filter

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7856150B2 (en) * 2007-04-10 2010-12-21 Arcsoft, Inc. Denoise method on image pyramid
US20120189051A1 (en) * 2008-04-11 2012-07-26 Yunfei Zheng Methods and apparatus for deblocking filtering of non-local intra prediction
CN101833753A (en) * 2010-04-30 2010-09-15 西安电子科技大学 SAR image de-speckling method based on improved Bayes non-local mean filter
CN102298773A (en) * 2011-09-19 2011-12-28 西安电子科技大学 Shape-adaptive non-local mean denoising method

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
ANTONI BUADES等: "A non-local algorithm for image denoising", 《COMPUTER VISION AND PATTERN RECOGNITION》 *
JIN WANG等: "fast non-local algorithm for image denoising", 《2006 IEEE INTERNATIONAL CONFERENCE ON IMAGE PROCESSING》 *
许光宇等: "有效保持细节特征的快速非局部滤波方法", 《计算机工程与应用》 *
许光宇等: "自适应的有效非局部图像滤波", 《中国国家图形学报》 *

Also Published As

Publication number Publication date
CN103679631B (en) 2018-01-23
WO2014044027A1 (en) 2014-03-27

Similar Documents

Publication Publication Date Title
US10880541B2 (en) Stereo correspondence and depth sensors
KR20170120639A (en) Deep Stereo: Running to predict new views from real-world images
CN103473799B (en) The method for dynamically processing of a kind of picture and device, terminal unit
WO2017020528A1 (en) Lane line recognition modeling method, apparatus, storage medium, and device, recognition method and apparatus, storage medium, and device
Banz et al. Real-time stereo vision system using semi-global matching disparity estimation: Architecture and FPGA-implementation
US9135710B2 (en) Depth map stereo correspondence techniques
JP4571987B2 (en) System and method for detecting features from an image of a vehicle
Ttofis et al. A low-cost real-time embedded stereo vision system for accurate disparity estimation based on guided image filtering
CN105654492A (en) Robust real-time three-dimensional (3D) reconstruction method based on consumer camera
US8717632B2 (en) Image processing method, image processing device and scanner
US7043091B2 (en) Method and apparatus for increasing spatial resolution of an image
CN104392045B (en) A kind of real time enhancing virtual reality system and method based on intelligent mobile terminal
CN105069754B (en) System and method based on unmarked augmented reality on the image
US8976168B2 (en) Mesh generation from depth images
CN103106688A (en) Indoor three-dimensional scene rebuilding method based on double-layer rectification method
US9208547B2 (en) Stereo correspondence smoothness tool
CN104363986B (en) A kind of image processing method and equipment
US10685201B2 (en) Method and system for recognizing location information in two-dimensional code
US9076234B2 (en) Super-resolution method and apparatus for video image
CN102859535A (en) Daisy descriptor generation from precomputed scale-space
CN102289803A (en) Image Processing Apparatus, Image Processing Method, and Program
CN101727580B (en) Image processing apparatus, image processing unit, and image processing method
Perri et al. Adaptive Census Transform: A novel hardware-oriented stereovision algorithm
JP6060632B2 (en) Method and apparatus for determining projection area in image
CN104183016B (en) A kind of construction method of quick 2.5 dimension building model

Legal Events

Date Code Title Description
PB01 Publication
C10 Entry into substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant